Console
The latest run on this page finished at 13:56 UTC on Tuesday 20 October. A run still going is not on this page yet.
Latest day
2026-10-20 ran 3 runs and published 205 of 240 planned items. 35 items failed.
3 runs are worth a look
Worst thing right now
26 feeds are resting, so nothing they carry reaches the digest. Each is asked again after 5 runs. On Pipelines.
Site size
46.3 MB of the 1 GB limit - room for about 386,000 more articles.
The days control above needs JavaScript, and so do the sections below - they draw rows a browser fetches. Their shape is 14 days.
This day ran in 3 runs, 1.51x apart on read speed. Hardware →
The article and score records are packed as far as 18 Oct 2026, so the 2 days after it are not shown yet.
A dashed line marks a day a recorded setting changed: the model, its settings, the prompt, the chat template, the runtime or the weights file. A dotted line marks a day the checker changed.
What the model did
What the model wrote for the day's own articles, how long it took, and what it got wrong.
Summaries today
322
- 18 Oct 2026
- Summaries today
- 322
Point at a day to read it.
-18% no target
Nothing changed about how the summaries are written inside these 14 days.
Marked "not sure"
28
- 18 Oct 2026
- Marked "not sure"
- 28
Point at a day to read it.
-30% of 322 summaries
How many of today's summaries we told you not to trust.
Nothing changed about how the summaries are written or checked inside these 14 days.
Numbers not in the article
5
- 18 Oct 2026
- Numbers not in the article
- 5
Point at a day to read it.
-44% of 322 summaries
The summary had a figure. The article did not.
Nothing changed about how the summaries are written inside these 14 days.
"Maybe" told as fact
51
- 18 Oct 2026
- "Maybe" told as fact
- 51
Point at a day to read it.
-14% of 322 summaries
The article said it might have happened. The summary said it did.
Nothing changed about how the summaries are written inside these 14 days.
Article read only in part
0
- 18 Oct 2026
- Article read only in part
- 0
Point at a day to read it.
The article was too long, so the machine read the start and stopped.
Nothing changed about how the summaries are written inside these 14 days.
Read only in part, as a percent
0%
- 18 Oct 2026
- Read only in part, as a percent
- 0%
Point at a day to read it.
The same articles, against the day's own count, so a busy day and a quiet one compare.
Nothing changed about how the summaries are written inside these 14 days.
Copied, not rewritten
14%
- 18 Oct 2026
- Copied, not rewritten
- 14%
Point at a day to read it.
+0% no target of 322 summaries
How much of a normal summary is lifted word for word.
Nothing changed about how the summaries are written inside these 14 days.
Time to write one
452 s
- 18 Oct 2026
- Time to write one
- 452 s
Point at a day to read it.
+4%
How long the machine takes on one article. The second figure is the articles it read only the start of.
Nothing changed about how the summaries are written inside these 14 days.
Model minutes
2841 min
- 18 Oct 2026
- Model minutes
- 2841 min
Point at a day to read it.
-33% no target
Nothing changed about how the summaries are written inside these 14 days.
Too long to send
0
- 18 Oct 2026
- Too long to send
- 0
Point at a day to read it.
The article and the instructions together did not fit, so the machine was never asked.
Nothing changed about how the summaries are written inside these 14 days.
Failed
73
- 18 Oct 2026
- Failed
- 73
Point at a day to read it.
+6%
Nothing changed about how the summaries are written inside these 14 days.
The same input gave a different summary
0
- 18 Oct 2026
- The same input gave a different summary
- 0
Point at a day to read it.
of 322 summaries
Set when a re-run of one item did not reproduce its own output.
Nothing changed about how the summaries are written inside these 14 days.
The cards show the newest measured day in this window. A missing reading is not measured, not zero.
Each card's newest day against its last 14 days.
Nothing acts on these numbers.
Show these figures day by day, over these 14 days
| Day | Summaries today | Marked "not sure" | Numbers not in the article | "Maybe" told as fact | Article read only in part | Read only in part, as a percent | Copied, not rewritten | Time to write one | Model minutes | Too long to send | Failed | The same input gave a different summary | Summary text that repeats itself | Summary sentences that follow one another | Article's most-used words the summary kept |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026-10-18 | 322 | 28 | 5 | 51 | 0 | 0% | 14% | 452 s | 2841 min | 0 | 73 | 0 | 0% | 0.30 | 50% |
| 2026-10-17 | 337 | 44 | 10 | 42 | 0 | 0% | 13% | 471 s | 3579 min | 0 | 91 | 0 | 0% | 0.30 | 50% |
| 2026-10-16 | 325 | 25 | 3 | 46 | 0 | 0% | 14% | 439 s | 2748 min | 0 | 75 | 0 | 0% | 0.31 | 50% |
| 2026-10-15 | 390 | 25 | 6 | 66 | 0 | 0% | 14% | 464 s | 3552 min | 0 | 71 | 0 | 0% | 0.31 | 50% |
| 2026-10-14 | 393 | 36 | 7 | 81 | 0 | 0% | 13% | 472 s | 4260 min | 0 | 56 | 0 | 0% | 0.31 | 50% |
| 2026-10-13 | 249 | 20 | 7 | 37 | 0 | 0% | 15% | 436 s | 2052 min | 0 | 264 | 0 | 0% | 0.30 | 50% |
| 2026-10-12 | 235 | 22 | 5 | 50 | 0 | 0% | 15% | 436 s | 2525 min | 0 | 22 | 0 | 0% | 0.32 | 55% |
| 2026-10-11 | 357 | 36 | 7 | 62 | 0 | 0% | 13% | 444 s | 3297 min | 0 | 36 | 0 | 0% | 0.30 | 50% |
| 2026-10-10 | 358 | 35 | 6 | 65 | 0 | 0% | 15% | 411 s | 3608 min | 0 | 32 | 0 | 0% | 0.31 | 50% |
| 2026-10-09 | 400 | 25 | 7 | 56 | 0 | 0% | 15% | 437 s | 3788 min | 0 | 41 | 0 | 0% | 0.30 | 50% |
| 2026-10-08 | 397 | 37 | 9 | 58 | 0 | 0% | 14% | 463 s | 4871 min | 0 | 64 | 0 | 0% | 0.31 | 50% |
| 2026-10-07 | 395 | 40 | 9 | 59 | 0 | 0% | 14% | 435 s | 4228 min | 0 | 69 | 0 | 0% | 0.30 | 50% |
Summary text that repeats itself The share of four-word runs the summary had already used. Up is the bad direction, and nothing bands it.
Summary sentences that follow one another How close each sentence sits to the one before it, from -1 to 1. A summary that says the same thing twice scores near the top, so it is a poor thing to judge one by.
Article's most-used words the summary kept Word for word, not by meaning: a summary saying "the chipmaker" where the article said "Nvidia" counts as a miss. A low reading is as often a good short summary as a poor one.
Set when a re-run of one item did not reproduce its own output.
Model tokens per second
How a model call ended
Which rule refused a reply
Why a summary was doubted
The columns run over these 14 days. The reasons stay separate, never added into one doubt count.
Each summary carries at most one published reason. The reason comes from its published item, not a second scoring rule.
The reason given most often is ""Maybe" told as fact": 610 summaries in these 14 days, or about one in every eight published.
- 20 Oct 2026, the newest day
- Numbers not in the article
- 2
- Does not match the article
- 18
- "Maybe" told as fact
- 28
Point at a day to read it. Left and Right step through them, Escape returns to the newest.
Nothing changed about how the summaries are written or checked inside these 14 days.
"Never checked" and "Left out the opening facts" are reasons the checker can give and did not give once in these 14 days, so they have no line on the chart.
Numbers not in the article; Does not match the article; Never checked; Left out the opening facts; "Maybe" told as fact.
The counts behind the reasons
| Day | Numbers not in the article | Does not match the article | Never checked | Left out the opening facts | "Maybe" told as fact |
|---|---|---|---|---|---|
| 2026-10-07 | 7 | 43 | 0 | 0 | 45 |
| 2026-10-08 | 7 | 51 | 0 | 0 | 46 |
| 2026-10-09 | 7 | 47 | 0 | 0 | 45 |
| 2026-10-10 | 5 | 46 | 0 | 0 | 48 |
| 2026-10-11 | 7 | 51 | 0 | 0 | 49 |
| 2026-10-12 | 5 | 33 | 0 | 0 | 30 |
| 2026-10-13 | 7 | 37 | 0 | 0 | 31 |
| 2026-10-14 | 7 | 46 | 0 | 0 | 65 |
| 2026-10-15 | 6 | 58 | 0 | 0 | 62 |
| 2026-10-16 | 3 | 43 | 0 | 0 | 43 |
| 2026-10-17 | 9 | 54 | 0 | 0 | 31 |
| 2026-10-18 | 5 | 46 | 0 | 0 | 41 |
| 2026-10-19 | 3 | 34 | 0 | 0 | 46 |
| 2026-10-20 | 2 | 18 | 0 | 0 | 28 |
The commonest reason against the rest. Each summary carries at most one recorded reason.